sh: Optimized readsl()/writesl() support.

Implement optimized copies of readsl()/writesl().

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

diff --git a/arch/sh/kernel/io.c b/arch/sh/kernel/io.c
index 71c9fde2fd9072517af4062168d720c2c3153b57..501fe03e3715770c484684f7b1fbecfb954d2010 100644 (file)
--- a/arch/sh/kernel/io.c
+++ b/arch/sh/kernel/io.c
@@ -61,6 +61,73 @@ void memset_io(volatile void __iomem *dst, int c, unsigned long count)
 }
 EXPORT_SYMBOL(memset_io);
 
+void __raw_readsl(unsigned long addr, void *datap, int len)
+{
+       u32 *data;
+
+       for (data = datap; (len != 0) && (((u32)data & 0x1f) != 0); len--)
+               *data++ = ctrl_inl(addr);
+
+       if (likely(len >= (0x20 >> 2))) {
+               int tmp2, tmp3, tmp4, tmp5, tmp6;
+
+               __asm__ __volatile__(
+                       "1:                     \n\t"
+                       "mov.l  @%7, r0         \n\t"
+                       "mov.l  @%7, %2         \n\t"
+#ifdef CONFIG_CPU_SH4
+                       "movca.l r0, @%0        \n\t"
+#else
+                       "mov.l  r0, @%0         \n\t"
+#endif
+                       "mov.l  @%7, %3         \n\t"
+                       "mov.l  @%7, %4         \n\t"
+                       "mov.l  @%7, %5         \n\t"
+                       "mov.l  @%7, %6         \n\t"
+                       "mov.l  @%7, r7         \n\t"
+                       "mov.l  @%7, r0         \n\t"
+                       "mov.l  %2, @(0x04,%0)  \n\t"
+                       "mov    #0x20>>2, %2    \n\t"
+                       "mov.l  %3, @(0x08,%0)  \n\t"
+                       "sub    %2, %1          \n\t"
+                       "mov.l  %4, @(0x0c,%0)  \n\t"
+                       "cmp/hi %1, %2          ! T if 32 > len \n\t"
+                       "mov.l  %5, @(0x10,%0)  \n\t"
+                       "mov.l  %6, @(0x14,%0)  \n\t"
+                       "mov.l  r7, @(0x18,%0)  \n\t"
+                       "mov.l  r0, @(0x1c,%0)  \n\t"
+                       "bf.s   1b              \n\t"
+                       " add   #0x20, %0       \n\t"
+                       : "=&r" (data), "=&r" (len),
+                         "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4),
+                         "=&r" (tmp5), "=&r" (tmp6)
+                       : "r"(addr), "0" (data), "1" (len)
+                       : "r0", "r7", "t", "memory");
+       }
+
+       for (; len != 0; len--)
+               *data++ = ctrl_inl(addr);
+}
+EXPORT_SYMBOL(__raw_readsl);
+
+void __raw_writesl(unsigned long addr, const void *data, int len)
+{
+       if (likely(len != 0)) {
+               int tmp1;
+
+               __asm__ __volatile__ (
+                       "1:                             \n\t"
+                       "mov.l  @%0+, %1        \n\t"
+                       "dt             %3              \n\t"
+                       "bf.s           1b              \n\t"
+                       " mov.l %1, @%4         \n\t"
+                       : "=&r" (data), "=&r" (tmp1)
+                       : "0" (data), "r" (len), "r"(addr)
+                       : "t", "memory");
+       }
+}
+EXPORT_SYMBOL(__raw_writesl);
+
 void __iomem *ioport_map(unsigned long port, unsigned int nr)
 {
        return sh_mv.mv_ioport_map(port, nr);

diff --git a/include/asm-sh/io.h b/include/asm-sh/io.h
index 377160b862950c3ea9641b6bc8b668aa1eec8f26..ed12d38e8c0082dda8db85613bf578f4819e3ef9 100644 (file)
--- a/include/asm-sh/io.h
+++ b/include/asm-sh/io.h
@@ -107,6 +107,9 @@
 #define __raw_writew(v, a)     __writew(v, (void __iomem *)(a))
 #define __raw_writel(v, a)     __writel(v, (void __iomem *)(a))
 
+void __raw_writesl(unsigned long addr, const void *data, int longlen);
+void __raw_readsl(unsigned long addr, void *data, int longlen);
+
 /*
  * The platform header files may define some of these macros to use
  * the inlined versions where appropriate.  These macros may also be
@@ -132,6 +135,9 @@
 # define writel(v,a)   ({ __raw_writel((v),(a)); mb(); })
 #endif
 
+#define writesl __raw_writesl
+#define readsl  __raw_readsl
+
 #define readb_relaxed(a) readb(a)
 #define readw_relaxed(a) readw(a)
 #define readl_relaxed(a) readl(a)